key: cord-0761786-w10pbi3d
authors: Ge, Y.; Zhang, W.; Liu, H.; Ruktanonchai, C. W.; Hu, M.; Wu, X.; Song, Y.; Ruktanonchai, N. W.; Yan, W.; Feng, L.; Li, Z.; Yang, W.; Liu, M.; Tatem, A. J.; Lai, S.
title: Effects of worldwide interventions and vaccination on COVID-19 between waves and countries
date: 2021-04-06
journal: nan
DOI: 10.1101/2021.03.31.21254702
sha: d7a38aac0644d909fe753c4299541f8d215fdaa2
doc_id: 761786
cord_uid: w10pbi3d

Worldwide governments have rapidly deployed non-pharmaceutical interventions (NPIs) to mitigate the COVID-19 pandemic, together with the large-scale rollout of vaccines since late 2020. However, the effect of these individual NPI and vaccination measures across space and time has not been sufficiently explored. By the decay ratio in the suppression of COVID-19 infections, we investigated the performance of different NPIs across waves in 133 countries, and their integration with vaccine rollouts in 63 countries as of 25 March 2021. The most effective NPIs were gathering restrictions (contributing 27.83% in the infection rate reductions), facial coverings (16.79%) and school closures (10.08%) in the first wave, and changed to facial coverings (30.04%), gathering restrictions (17.51%) and international travel restrictions (9.22%) in the second wave. The impact of NPIs had obvious spatiotemporal variations across countries by waves before vaccine rollouts, with facial coverings being one of the most effective measures consistently. Vaccinations had gradually contributed to the suppression of COVID-19 transmission, from 0.71% and 0.86% within 15 days and 30 days since Day 12 after vaccination, to 1.23% as of 25 March 2021, while NPIs still dominated the pandemic mitigation. Our findings have important implications for continued tailoring of integrated NPI or NPI-vaccination strategies against future COVID-19 waves or similar infectious diseases.

to avoid further resurgences before herd immunity can be achieved 4 . It is critical to understand the role of different NPIs and initial vaccination efforts to reduce COVID- 19 transmission, before and after vaccine rollouts, thereby tailoring effective and integrated NPI-vaccination strategies for future COVID-19 waves.

The effectiveness of NPIs on pandemic mitigation had been shown by previous studies that mostly focused on the first wave of the pandemic before July 2020 5, [8] [9] [10] [11] [12] , with limited analysis of subsequent waves, regional diversity and integrated NPI-vaccination efforts. The implementation of NPIs in the first wave had, to some degree, changed human knowledge and perceptions, behaviours and responses to mitigate the outbreaks [13] [14] [15] [16] [17] . Whether NPI effectiveness increases with adherence or decreases with fatigue in the subsequent waves remains unclear. Additionally, the effects of NPIs may vary across countries with different country characteristics, such as health capacity, residential population density, aging ratio, humidity and air temperature 18, 19 . The potential differences in NPI effectiveness across continents are rarely discussed in existing global analyses 12 . Moreover, vaccination is the most promising approach to lead the way out from this pandemic. However, the uneven distribution and allocation of vaccine rollout among countries and population groups might hinder the way to herd immunity 20 . Modelling studies have been conducted to simulate the combining effects of vaccination and NPIs for COVID-19 under various scenarios 5, 21, 22 . However, it is critically needed to understand how vaccination integrated with NPIs reduces COVID-19 transmission in the real world since the rollout of vaccines across multiple nations.

In this study we estimated the effects of individual NPIs and vaccination by identifying their contributions to the decay ratio of COVID-19 infections across waves and countries after the implementation of these measures. We used databases of global comparable outcomes, covering epidemiological 23 , intervention policy 24 , environmental, and demographic data in133 countries, territories and areas, from the earliest available dates to 25 ), in the presence, absence and intensity change of these interventions. The decay ratio was defined as a percentage of reduction in the baseline growth rate by the instantaneous growth rate. In addition to interventions, there were many other factors (e.g., the transmissibility of new variants and the variation of case diagnosis and reporting) that might affect the transmission of COVID-19 over time. Therefore, the baseline growth rates in different waves and . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) countries were assumed as the mean of the top three highest instantaneous, weekly growth rates in the corresponding wave and country. The instantaneous growth rate of transmission at each point of time was calculated as the current weekly number of new infections over the infections in the previous week. We used the decay ratio directly derived from the reported case data, rather than the reproduction number (Rt) 10, 12 , to avoid introducing the uncertainty of estimating Rt over time 25 . It should be noted that we estimated the relative effects of individual measures, while their combined effectiveness should be higher than individual effects, but not linearly accumulated (see Method).

We modelled NPI effects over time without assuming a functional relationship between effectiveness over time, which allows for variable community responses to the variation of each intervention. The effects of each NPI and vaccination with same intensity were assumed to be constant across countries in our model for each single estimation, and then decomposed for each country and week according to the corresponding decay ratio, intervention timing and intensity. The NPI with different intensities was modelled by the same effect parameter. The spatial variations in NPI and vaccination effectiveness across countries were controlled by employing the country-specific characteristics, including health capacity, residential population density, aging ratio, humidity and air temperature. All the estimations were performed by Markov chain Monte Carlo (MCMC). The reliability of our model was assessed by the cross-validation for overall intervention effects. Sensitivity analyses were also performed to assess model robustness in terms of our assumptions. More details on models and covariates can be found in Methods and SI.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted April 6, 2021. ;  We estimated that three NPIs had substantial effects (>10%) on mitigating COVID-19 transmission in general (Fig. 1 represents the decay ratio in the COVID-19 infection rate in 133 studied countries, territories and areas. The 5th, 25th (Q1), 50th (median), 75th (Q3), and 95th percentiles of estimates are presented, respectively. The uncertainty intervals of NPI effectiveness refer to the variance over the corresponding data context.

The efficacy of NPIs varied across waves. In the first wave, the most effective NPIs were gathering restrictions (median 27.83%, IQR 25.60 -29.97%), facial coverings (16.78%, 15.82 -17.74%), and school closures (10.08%, 6.70 -12.91%). In the second wave, the efficacy of facial coverings surged to be the top-ranked one (30.04%, 28.14 -31.94%). Another significant rise is the effect of international travel restrictions, from limited in Wave 1 (0.96%, 0.03 -3.55%) to moderate in Wave 2 (9.22%, 7.07 -11.25%). Meanwhile, the effects of gathering restrictions and school closures declined to 17.51% (IQR 15.08 -19.82%) and 0.06% (0.00 -0.67%) in the second wave, respectively. In both waves, workplace closures, public transport restrictions and movement restrictions presented limited effects (< 5%) in reducing the transmission.

Our analyses also revealed that the impact of individual non-pharmaceutical measures represents the decay ratio in COVID-19 infection rate. The 5th, 25th (Q1), 50th (median), 75th (Q3), and 95th percentiles of estimates are presented, respectively. The uncertainty intervals of NPI effectiveness refer to the variance over the corresponding spatiotemporal extent. A full list of countries and the corresponding time frames of different waves for each group can be found in SI Table  E1 -E4. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted April 6, 2021. 

We compared the effects of vaccination and NPIs after the first-dose vaccination in Our results presented NPI effectiveness along both spatial and temporal scales, and this study was the first impact assessment of integrating worldwide COVID-19

interventions and the vaccination in the real world, to our knowledge. These findings are crucial for continued tailoring and implementation of NPI strategies to mitigate COVID-19 transmission among future waves (e.g., as a result of new variants) or similar emerging infectious diseases, such as pandemic influenza.

Preliminary data showed vaccines could significantly reduce the severity of infections in older people 26 , and our results also showed that vaccines have an increasing effect to reduce SARS-CoV-2 transmission in the whole population, while vaccination alone was still insufficient to fully contain the coronavirus spread, for the time being, considering the vaccinated population ratio in most countries below 10% by 25 March 2021. Mass vaccination is needed to confer broad protection to the coronavirus, through reducing the unevenness of vaccine distribution among regions and groups 20 . However, the efficacy of vaccines and herd immunity might be . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted April 6, 2021. ; https://doi.org/10.1101/2021.03.31.21254702 doi: medRxiv preprint undermined due to the emergence of new variants of SARS-CoV-2, the wane of infection-associated immunity over time, and the changing attitudes and behaviours on vaccination, even with vaccine roll-out in full force 20 . Therefore, it is necessary to maintain the implementation of target and effective NPIs and closely monitor the changing efficacy of NPIs and vaccines across waves and countries for local intervention design.

We found that gathering restrictions and facial coverings significantly changed the pandemic trajectory in both waves, and gathering restrictions include both gathering cancellation 27 and closure of non-essential businesses [28] [29] [30] . The significant effects of these measures might be due to the virus most commonly spread through droplets or aerosols among people who were in close contact 31 (Fig. 1) . Countries that quickly placed border controls might have reduced the seeding of the coronavirus between countries, but international travel restrictions cannot prevent local transmission at the community level in countries where the virus had already been introduced 35, 36 . Previous models suggest that unless community-level transmission is reduced by no less than 50%, a reduction in 90% of international travel to and from epidemic centres might only modestly affect the epidemic trajectory 37 . The small effect of international travel restrictions in the first wave might be explained by the late implementation of this measure across countries.

The increasing role of international travel restrictions observed in this study might be also due to increasing control efforts at community level which occurred during the second wave.

After controlling for local contextual confounders in our models, we observed variations in the efficacy of interventions across regions. In this study, we divided 133 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted April 6, 2021. ;  countries into four groups based on mortality and morbidity, implicitly related to testing rate. The testing rate was highly correlated with the mortality and morbidity (see SI Table C1 ). Regardless, socio-economic, cultural and political characteristics could also affect the implementation and effectiveness of NPIs locally 31 Overall, the disclosure of epidemic, publicized responses and the COVID-19 vaccination data allows us to estimate and compare the cross-wave effects of public health measures at both global and regional scales. Our work provides a quantitative basis and approach to explore historic spatial-temporal heterogeneity in the effectiveness of individual NPIs, integrating vaccinations. The continued pandemic burden across the globe and the non-decisive efficacy of the vaccination suggests the NPI implementation continues to be a priority for many countries, even with a full force vaccine rollout in the early stage of the vaccination era 5, 20, 38 .

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted April 6, 2021. ; https://doi.org/10.1101/2021.03.31.21254702 doi: medRxiv preprint

Epidemiological data. The daily number of confirmed cases reported by country were obtained from the COVID-19 Data Repository by the Center for Systems Science and Engineering (CSSE) at Johns Hopkins University (JHU) 23 . To remove the influence of outliers and the fluctuation caused by the day-of-week effect, we smoothed daily case counts with the Gaussian kernel and adjusted them for the infection-to-confirmation reporting delay then. First, we set all negative case numbers to zero, as negative values in cases could sometimes appear when a country corrected historical data, because it had previously overestimated the number of cases 39 .

Second, we smoothed case data by calculating the rolling average using a Gaussian window with a standard deviation of 2 days, truncated at a maximum window of 15 days 27 . Third, we adjusted the reporting data by subtracting 12 days 15 from the reporting date in the first wave, accounting for the delay from infection to reporting.

With respect to the second, even the third, wave, we slightly reduced the reporting delay to 10 days because of the potential increasing testing and diagnosis capacity.

Sensitivity analyses were also conducted to assess the impact of different reporting delays on our estimates (see sensitivity analysis section in Methods).

Intervention policy data. We generated seven non-pharmaceutical measures from the nine NPIs (i.e. school closures, workplace closures, public events cancellations, gathering restrictions, public transport closures, stay-at-home orders, internal movement restrictions, international travel restrictions, and facial coverings), collated by the Oxford COVID-19 Government Response Tracker (OxCGRT) 24 . The intensity of the nine considered NPIs policies is scaled into discrete values between 0 to 1, . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted April 6, 2021. ; https://doi.org/10.1101/2021.03.31.21254702 doi: medRxiv preprint where 0 represents an absence of the NPI and 1 represents the corresponding maximum intensity. The intensity of school closures is corrected as 1 during public and school holidays 40 . We processed public events cancellations and gathering restrictions into a single NPI (i.e., gathering restrictions) for each country in each day by their mean intensity, as the two NPIs documented in OxCGRT were highly collinear in terms of their timing and intensity of implementation across the 133 study countries. We also integrated stay-at-home orders and internal movement restrictions into movement restrictions for the same reason in the same way for each country.

Vaccination data. The COVID-19 vaccination data used in this study was obtained from the Our World in Data 41 . They regularly updated the first and second doses administered and daily vaccination rates at national scale from official sources in 93 countries as of 25 March 2021. We analysed the vaccination effect in 63 countries whose highest daily confirmed cases exceeded 100. The induced antibody response and immunity might sufficiently prevent SARS-CoV-2 infections since Day 12 after receiving the first dose 42 . Therefore, we adjusted the vaccination rates for the first dose administered to be rolled forward for 12 days, to account for the delay from vaccination to the generation of sufficient protective immunity.

Environmental and demographic covariates. To control for country-specific confounders in the estimates of intervention effectiveness varied across countries, we also assembled population density, aging ratio, health capacity index, air temperature, and humidity for all these 133 study countries. Within each country, population density (per square kilometre) was the ratio of the total population over the corresponding built-up area in 2014 43 . The total and age-grouping population data in . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted April 6, 2021. ; https://doi.org/10.1101/2021.03.31.21254702 doi: medRxiv preprint 2019 were obtained from the United Nations to calculate the aging ratio (> 65 year old) among populations 44 . Health capacity index was the arithmetic average of the five indices, including i) prevent, ii) detect, iii) respond, iv) enabling function, and v) operational readiness, developed to characterize the health security capacities in the context of the COVID-19 outbreak 45 . Air temperature and humidity were derived from the Global Land Data Assimilation System 46 .

To further remove the day-of-week effect among case testing, diagnosis, and data reporting, all data used in this study were assembled and aggregated into a weekly dataset. The correlations between each two covariates were given to show their collinearity. The studied countries were selected by being documented in every dataset of epidemiological data, intervention policy data and environmental and demographic covariates. The details of data collection and processing are further provided in the Supplementary Information.

Waves. The inequality in pandemic development across worldwide countries has led some countries to confront more than one COVID-19 wave 47, 48 . Based on the smoothed daily case data, we defined an epidemic wave in each country as below. In a period of three or more consecutive weeks for a country, if the daily numbers of cases in this period all exceeded 5% of the maximum daily number of cases in 2020 in this country, these weeks were considered to constitute an epidemic wave. The first and last days of the period were the start and end of the corresponding wave, respectively. However, considering that the first wave of this pandemic in most countries started from low-level community transmission caused by imported cases, . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted April 6, 2021. ; https://doi.org/10.1101/2021.03.31.21254702 doi: medRxiv preprint we adjusted the start date of the first wave to the first date: i) when the number of daily cases exceeded 10 cases, for countries where the maximum number of daily new cases in the first wave were less than or equal to 300 cases; or ii) when the number of daily cases exceeded 20 cases, for other countries where their maximum daily cases in the first wave were greater than 300 cases. The details and full lists of waves by country can be found in SI. We focused on the first and second waves in the main text, and the results of the third waves that were only identified in a few countries are provided in SI.

Regional stratification. The reported COVID-19 morbidity and mortality could vary substantially in the study countries, based on the released epidemiological data. We investigated the spatial heterogeneity in NPIs effectiveness by dividing 133 countries into four country groups based on their morality, mortality and geographical proximity. Among the four groups, the grading thresholds for high morbidity and mortality were determined according to the principle of "small variance within groups and large variance between groups" 49, 50 . Thresholds of 1800 per 100,000 persons for morbidity and 40 per 100,000 persons for mortality were chosen to select countries with both high morbidity and high mortality. Considering the geographical proximity between countries, Asian countries and African countries were assigned into two separate groups. A full list of countries in each group and the corresponding time frame of different waves of COVID-19 can be found in SI is the . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted April 6, 2021. ; https://doi.org/10.1101/2021.03.31.21254702 doi: medRxiv preprint instantaneous growth rate. We adopted a general linear formula 9, 10, 35 linking NPIs to the pandemic evolution. That is, change across waves and country groups, but relatively stable for countries within the same group. The . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted April 6, 2021. differences in NPIs effectiveness across country groups were assumed to be too large to be controlled by sociodemographic factors. We first evaluated the overall effectiveness of NPIs before the vaccination era. In addition to the overall NPIs effectiveness, we also evaluated respective NPIs effects in the first and second waves for each country group separately to show the potential large spatiotemporal diversity. Model validation. The reliability of our models and corresponding results were evaluated by the leave-forty-countries-out cross validation. We first calibrated our . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted April 6, 2021. ; https://doi.org/10.1101/2021.03.31.21254702 doi: medRxiv preprint model using 93 randomly selected countries to estimate the overall NPIs effects in both the first and second waves. Then, we derived the instantaneous growth rates through the estimated NPIs overall effects for the remaining 40 countries in terms of their implemented NPIs. We used mean square error, ranging from 0 to infinite with 0 represents the perfect prediction ability, to assess the difference between the predicted instantaneous growth rates and the corresponding empirical instantaneous growth rates. We repeated this procedure 50 times, where the average mean square error was 1.01. Table B2 , representing three scenarios with smaller and larger default parameter settings. The differences of NPI effects among three waves were tested using a Wilcoxon signed-rank test, a non-parametric statistical hypothesis test for comparing NPIs effects between pairs of the three waves. Moreover, we repeated the estimations twice for NPIs effectiveness with default setting, except for the initial full infection rate, as the highest growth rate and the mean of the top five highest growth rates, respectively, of the confirmed COVID-19 new cases in the corresponding wave.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted April 6, 2021.

Using an R package, rstan 51 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted April 6, 2021. ; https://doi.org/10.1101/2021.03.31.21254702 doi: medRxiv preprint

World Health Organization. WHO Coronavirus (COVID-19) Dashboard

Non-pharmaceutical interventions during the COVID-19 pandemic: A review

Epidemiological and evolutionary considerations of SARS-CoV-2 vaccine dosing regimes

Integrated COVID-19 response in the vaccination era

Integrated vaccination and physical distancing interventions to prevent future COVID-19 waves in Chinese cities

Preliminary risk analysis of the spread of new COVID-19 variants from the UK, South Africa and Brazil

Viral mutations may cause another 'very, very bad'COVID-19 wave, scientists warn

Assessing the influence of climate on wintertime SARS-CoV-2 outbreaks

Effect of non-pharmaceutical interventions to contain COVID-19 in China

Estimating the effects of non-pharmaceutical interventions on COVID-19 in Europe

An investigation of transmission control measures during the first 50 days of the COVID-19 epidemic in China

The temporal association of introducing and lifting nonpharmaceutical interventions with the time-varying reproduction number (R) of SARS-CoV-2: a modelling study across 131 countries

Impact of Online Information on Self-Isolation Intention During the COVID-19 Pandemic: Cross-Sectional Study

Impact of the COVID-19 Epidemic on Lifestyle Behaviors and Their Association With Subjective Well-Being Among the General Population in Mainland China: Cross-Sectional Study

Inferring the effectiveness of government interventions against COVID-19

Knowledge and perceptions of COVID-19 among the general public in the United States and the United Kingdom: a cross-sectional online survey

Public Perceptions and Attitudes Towards COVID-19 Non-Pharmaceutical Interventions Across Six Countries: A Topic Modeling Analysis of Twitter Data (Preprint)

Using social and behavioural science to support COVID-19 pandemic response

The effect of large-scale anti-contagion policies on the COVID-19 pandemic

Five reasons why COVID herd immunity is probably impossible

Vaccination and non-pharmaceutical interventions for COVID-19: a mathematical modelling study. The Lancet Infectious Diseases

The Joint Impact of COVID-19 Vaccination and Non-Pharmaceutical Interventions on Infections, Hospitalizations, and Mortality: An Agent-Based Simulation

An interactive web-based dashboard to track COVID-19 in real time

A global panel database of pandemic policies (Oxford COVID-19 Government Response Tracker)

Range of reproduction number estimates for COVID-19 spread

New data show vaccines reduce severe COVID-19 in older adults

Ranking the effectiveness of worldwide COVID-19 government interventions

Impact of non-pharmaceutical interventions on documented cases of COVID-19

Predicting spatial and temporal responses to non-pharmaceutical interventions on COVID-19 growth rates across 58 counties in New York State: A prospective event-based modeling study on county-level sociological predictors. JMIR public health and surveillance

Preparing for a future COVID-19 wave: insights and limitations from a data-driven evaluation of non-pharmaceutical interventions in

Stopping the Spread of COVID-19

COVID-19 and public transportation: Current assessment, prospects, and research needs

Universal masking in the United States: The role of mandates, health education, and the CDC

Lockdown strategies, mobility patterns and covid-19

Effect of internationally imported cases on internal spread of COVID-19: a mathematical modelling study

Impact of international travel and border control measures on the global spread of the novel 2019 coronavirus outbreak

The effect of travel restrictions on the spread of the 2019 novel coronavirus (COVID-19) outbreak

Reduction in mobility and COVID-19 transmission

A cross-country database of COVID-19 testing

Global Public and School Holidays

COVID-19) Vaccinations

Safety and efficacy of the BNT162b2 mRNA Covid-19 vaccine

GHSL data package

World population prospects 2019: highlights. Department of Economic and Social Affairs

Health security capacities in the context of COVID-19 outbreak: an analysis of International Health Regulations annual report data from 182 countries

The global land data assimilation system

Africa Succeeded Against COVID-19's First Wave, but the Second Wave Brings New Challenges

Modelling the impact of testing, contact tracing and household quarantine on second waves of COVID-19

Geodetector: Principle and prospective

A measure of spatial stratified heterogeneity

RStan: the R interface to Stan

Ethical clearance for collecting and using secondary data in this study was granted by the institutional review board of the University of Southampton (No. 61865). All data were supplied and analysed in an anonymous format, without access to personal identifying information.

The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding authors had full access to all the data in the study and had final responsibility for the decision to submit for publication. The views expressed in this article are those of the authors and do not represent any official policy.

The authors declare no competing interests.